The Relationship Between Short-Term Intraindividual Variability and Longitudinal Intraindividual Cognitive Change in Older Adulthood:
Covariation and Prediction of Change by
Allison Anne Marie Bielak B. A., University of Winnipeg, 2002 M. Sc., University of Victoria, 2004
A Dissertation Submitted in Partial Fulfillment of the Requirements for the Degree of
DOCTOR OF PHILOSOPHY
in the Department of Psychology
© Allison Anne Marie Bielak, 2008 University of Victoria
All rights reserved. This dissertation may not be reproduced in whole or in part, by photocopy or other means, without the permission of the author.
The Relationship Between Short-Term Intraindividual Variability and Longitudinal Intraindividual Cognitive Change in Older Adulthood:
Covariation and Prediction of Change by
Allison Anne Marie Bielak B. A., University of Winnipeg, 2002 M. Sc., University of Victoria, 2004
Supervisory Committee
Dr. David F. Hultsch, Supervisor (Department of Psychology)
Dr. Esther Strauss, Departmental Member (Department of Psychology)
Dr. Michael A. Hunter, Departmental Member (Department of Psychology)
Dr. Ryan Rhodes, Outside Member
Supervisory Committee
Dr. David F. Hultsch, Supervisor (Department of Psychology)
Dr. Esther Strauss, Departmental Member (Department of Psychology)
Dr. Michael A. Hunter, Departmental Member (Department of Psychology)
Dr. Ryan Rhodes, Outside Member
(School of Exercise Science, Physical and Health Education)
Abstract
This dissertation presents two studies of intraindividual variability in a longitudinal context to further explore the relationship between short-term intraindividual variability and longer-term cognitive change in older adults. A sample of 304 community-dwelling older adults initially aged 64-92 years completed between 1 to 6 waves of annual testing over a 5-year period. Participants completed an extensive battery of accuracy- and latency-based tests covering a wide range of cognitive complexity. The first study
addressed the longitudinal nature of intraindividual variability over 3 years. Group-based increases in inconsistency were limited to the latter half of older adulthood (i.e., 75 years and older), but there were significant individual differences across the entire sample. The covariation relationships between change in cognition and change in inconsistency were significant across the one-year interval, and found to remain stable across both time and older age. For each additional unit increase in intraindividual variability, participants’ cognitive performance correspondingly declined. The strength of the coupling
reasoning, and processing speed, and variability based on moderately and highly complex tasks provided the strongest prediction. Building on these results suggesting that
intraindividual variability is highly sensitive to even subtle changes in cognitive ability, the second study addressed the capacity of intraindividual variability to predict cognitive ability and other meaningful change outcomes 5 years later. Inconsistency at Wave 1 was particularly sensitive to changes reflecting the early behavioural characteristics of
dementia, including episodic memory ability, cognitive status, and attrition. In each case, greater inconsistency at baseline was associated with a greater likelihood of being in a maladaptive group 5 years later. Mean rate of responding was a comparable predictor of change in most instances, but differences emerged according to the complexity of their derived tasks. Variability based on moderate to high cognitively challenging tasks appeared to be the most sensitive to longitudinal changes in cognitive ability, and was uniquely predictive of the rate of attrition compared to neuropsychological tasks. These findings are promising of the potential utility and applicability of intraindividual
variability in understanding and predicting intraindividual cognitive change in older adulthood.
Table of Contents Page Title Page………..i Supervisory Committee………ii Abstract………...iii Table of Contents ...v
List of Tables ... vii
List of Figures ... viii
Acknowledgements...ix
Introduction...1
Why is Intraindividual Variability Important?...1
Various Definitions of Intraindividual Variability...2
Reliability of Intraindividual Variability...5
Stability of Intraindividual Variability...5
Intraindividual Variability and Various Stable Characteristics...8
Intraindividual Variability in Other Domains...10
Neurological Correlates of Intraindividual Variability...12
Age Differences and Changes in Intraindividual Variability...20
Cross-sectional Differences Across the Lifespan...20
Longitudinal Changes with Age...23
Overview of Studies...28 Study 1...29 Research Questions...32 Method...34 Participants...34 Procedure...36 Measures...37 Data Preparation...41 Results...46
Relation of Wave 1 Inconsistency to Wave 3 Cognitive Ability...46
3-Year Change in Cognitive Ability...51
3-Year Change in Intraindividual Variability...56
Association Between 3-Year Change in Cognitive Performance and Inconsistency.59 Variations in the Covariation Relationship due to Cognitive Complexity...67
Discussion...69
Predicting Level of Cognitive Ability 3 Years Later...69
3-Year Change in Cognition and Inconsistency...70
Cognition...70
Inconsistency...72
Covariation of Cognition and Inconsistency...74
Limitations and Future Directions...80 Study 2...82 Research Questions...85 Method...88 Participants...88 Procedure...89 Measures...89
Construction of Outcome Groups...91
Data Preparation...100
Results...104
a) Level of Cognitive Performance at Wave 6 ...105
b) Change in Cognitive Performance ...114
c) MMSE Change ...123
d) Cognitive Status Change...125
e) Attrition...128
Inconsistency and Neuropsychological Tests Predicting Rate of Attrition...135
Discussion...137
Cognitive Level...137
Cognitive Change...138
MMSE Change...141
Cognitive Status Change...142
Attrition...145
Across the Various Outcomes...149
Intraindividual Variability Versus Intraindividual Mean...151
Differences Due to Task Complexity...154
Intraindividual Variability versus Neuropsychological Tests in Predicting Attrition ...154
Limitations and Future Directions...155
General Discussion...157
List of Tables
Page Table 1: Univariate Test Results of Year 1 ISD Composites Predicting Year 3
Cognitive Test Score………..……….………...48 Table 2: Model Fit Using Various Time Metrics for a Sample of Outcome Measures...50 Table 3: 3-Year Change in Cognitive Performance……….53 Table 4: 3-Year Change in Intraindividual Variability………...…...57 Table 5: Rate of Cognitive Change as a Function of Change in Inconsistency,
Controlling for Time in Study
a) Digit Symbol, Letter Series, Word Recall………...…63 b) Similarities, Vocabulary……….……….…....64 Table 6: Cell Sizes for each Change Status per Cognitive Outcome Change Group…...94 Table 7: Univariate Test Results of Year 1 ISD Composites and Age Group Predicting
Year 6 Cognitive Test Score………...……….106 Table 8: Hierarchical Regressions of ISD and IM Composites Predicting Year 6
Cognitive Test Score………...………..…...108 Table 9: Univariate Test Results of Year 1 IM Composites and Age Group Predicting
Year 6 Cognitive Test Score………...……….109 Table 10: Odds Ratios of Year 1 ISD Composites Individually Predicting Cognitive
Test Change Groups………111
Table 11: Odds Ratios of Year 1 IM Composites Individually Predicting Cognitive Test Change Groups………..…..112 Table 12: Odds Ratios of Year 1 ISD and IM Composites Individually Predicting
CIND Change Groups………..………...113 Table 13: Model Statistics of Year 1 ISD Composites and Age Group Predicting Rate
of Attrition, Controlling for Other Predictors……….131
Table 14: Model Statistics of Year 1 IM Composites and Age Group Predicting Rate of Attrition, Controlling for Other Predictors……….134
Table 15: Hierarchical Regressions of Basic ISD and Neuropsychological Tests
Predicting Rate of Attrition………...136 Table 16: Reported Reasons for Attrition ………...…...146
List of Figures
Page Figure 1: Word Recall Change Over Time in Study for a Sample of Participants….….54 Figure 2: Complex ISD Change Over Time in Study for a Sample of Participants……58 Figure 3: Estimated Individual Regression Lines Describing Change in Digit Symbol
and Change in Basic ISD at any Time in Study………...65 Figure 4: Possible Change Trajectories in CIND Classification over 5 Years…………98 Figure 5: Cumulative Survival Function of Attrition by Time in Study………129 Figure 6: Cumulative Survival Function of Attrition by Time in Study According to
Acknowledgements
Although it feels surreal to be at the end of my doctoral training, I know a lot of people saw this moment clearly all along. My first and foremost thanks go to my
supervisor, Dr. David Hultsch, who expertly guided my progress and patiently helped me to develop my scholastic talents into what they are today. Thanks to his mentorship and advice, I know my future academic career will be a promising one. My sincere gratitude also goes to Dr. Esther Strauss, Dr. Michael Hunter, and Dr. Roger Dixon, who have each played a special supportive role in helping me develop and learn over the years. As a student who was initially unsuccessful in securing external funding, I am extremely grateful for the financial support offered to me by the Michael Smith Foundation for Health Research, the BC Medical Services Foundation, the Canadian Institutes of Health Research, and various local University donors. The completion of the statistical analyses herein and particularly my understanding of them would not have been possible without the guidance of Dr. Stuart MacDonald, who was extremely generous with his time, patience, and inspiring words. I am also appreciative to the staff at the Victoria
Longitudinal Study, Arlene from Project MIND, and the committed participants of both longitudinal studies. I must acknowledge those around me day in and day out, including my wonderful office and lab mates over the years, and friends who were always willing to discuss ideas and share experiences and advice. Finally, to those closest to me, I cannot thank you enough. My parents were my long-distance cheering squad, and my boyfriend Mark’s unwavering belief in my abilities and constant encouragement kept me on time and pushing forward. Thank you for always believing in me.
Why is Intraindividual Variability Important?
Researchers investigating cognitive functioning in adulthood tend to examine adults’ mean level of performance on various cognitive tasks. Such measurement allows easy comparison across individuals of different age groups, or calculation of within-person longitudinal change over time. However, this type of measurement consequently assumes that individuals’ performance is also stable in the short-term, or from one
moment to the next. In fact, there are theoretical arguments against this assumption. The life-span development perspective states that an individual is in constant fluctuation as a result of living in a dynamic environment (Nesselroade, 1991), and that stability is only a temporary break from the ongoing variations in functioning (Nesselroade & Featherman, 1997). Therefore, the study of developmental change has to include and expect both variability and stability in development (Nesselroade & Featherman, 1997). These arguments suggest that research investigating the mean level of performance in older adulthood has only been considering part of the story.
Nesselroade (1991) clarified these two developmental distinctions further by distinguishing between two types of change: Intraindividual change is defined as changes that occur relatively slowly and over a relatively long period of time (e.g., year-to-year), and result in long-lasting changes in an individual (e.g., developmental change, learning of skills). On the other hand, intraindividual variability is defined as changes that occur relatively quickly and over relatively short time frames (e.g., moment-to-moment; week-to-week), and are temporary or reversible. Examples include shifts in mood, and
fluctuations in physical or cognitive performance. A further distinction between the two types of change is statistical, in that intraindividual changes represent systematic change,
whereas intraindividual variability is unsystematic. For example, on a reaction time task a person’s intraindividual change would be represented by their beta weight, or statistical change over time, but their intraindividual variability would be represented as the
variation in performance around their own personal regression line. However, a dynamic relationship between the two change types likely exists, where short-term variations play a role in the occurrence of long-term developmental change (Nesselroade, 1991).
Therefore, changes in the distributions of intraindividual variability may be indicators of impending longitudinal intraindividual change (Hultsch, Strauss, Hunter, & MacDonald, 2008). This dissertation evaluated these hypotheses by examining whether
intraindividual variability changes with and is predictive of later longitudinal outcomes. Various Definitions of Intraindividual Variability
Although intraindividual variability is defined as short-term but reversible change, further distinctions of the concept can be made along various dimensions. First, the time frame against which short-term change occurs is particularly important. For example, it is possible for transient change to exist across moments, trials, days, weeks, or even months. This diversity is evident in the research literature, as some studies have focused on particularly rapid change across trials of a reaction time (RT) task (e.g., Hultsch, MacDonald, & Dixon, 2002), some have focused on day-to-day fluctuations (e.g., Sliwinski, Smyth, Hofer, & Stawski, 2006), and others have investigated performance fluctuations across weeks (e.g., Eizenman, Nesselroade, Featherman, & Rowe, 1997). A handful of studies have even compared variability measurements based on different time frames. For example, Hultsch, MacDonald, Hunter, Levy-Bencheton, and Strauss (2000), Nesselroade and Salthouse (2004), and Rabbitt, Osman, Moore, and Stollery (2001), all found positive correlations between trial-based variability and session-based variability;
individuals who were more variable from trial-to-trial were also more variable from session-to-session. However, Rabbitt and colleagues noted that the trial-based intraindividual variability did not completely account for the session-based
intraindividual variability, indicating that the underlying causes of variability may be slightly different depending on their derived time frame.
This leads into the second important distinction of intraindividual variability; there are varying potential sources of influence or causes. For example, weekly variability is more likely to be influenced by external factors like stress or fatigue than variability occurring across the trials of a task. Rather, the causes of rapid variability across trials are more likely to be endogenous, or derived from causes within the individual, such as the connectivity of neural networks or the efficiency of neurotransmitters. In fact, Martin and Hofer (2004) described a range of possible causes for variability based on its time frame, from attentional lapses for moment-based measures, fatigue, motivation, and order effects for session comparisons, and environmental causes, physical health, or practice effects for daily or weekly fluctuations. Therefore, it is important that the type of intraindividual variability being used reflects the source of influence one hopes to investigate.
A third distinction involves the scope of intraindividual variability (Li, Huxhold, & Schmiedek, 2004). Is the variability measured as fluctuations in performance on a single task (univariate), or is the variability defined as changes in the organization of abilities while performing a number of tasks (multivariate)? Li and colleagues noted that variations on a single task most likely represent endogenous mechanisms such as
paradigm most likely represents switches in resource allocation, compensation, and organization.
Finally, intraindividual variability can be viewed as either adaptive or
maladaptive, in that it is associated with positive or negative outcomes. Siegler (1994) described intraindividual variability in childhood as critical in promoting cognitive change, suggesting that the fluctuations act as evidence of trying different strategies for completing a task. He found that the trial immediately preceding the discovery of a new strategy and the trial on which the discovery was made were more variable. Therefore, in situations of learning, intraindividual variability may be adaptive. Similarly, Allaire and Marsiske (2005) found intraindividual variability was positively correlated with cognitive performance in older adults. According to Li, Huxhold et al. (2004), the adaptability of greater intraindividual variability appears to depend on a few key factors. Variability in responding is adaptive if the optimal way to perform the task is unknown, thus requiring one to experiment with various strategies, and there is consequently substantial room for growth and improvement on the task. This situation is more likely to occur on tasks that focus on accuracy rather than speed. However, there are limitations to the adaptive nature of intraindividual variability; once the level of learning asymptotes and peak performance is reached, continued intraindividual variability in responding is deemed to be
maladaptive. Alternatively, intraindividual variability in responding on tasks that have little room for improvement (i.e., performance is already near ceiling), and where the current method of completing the task is appropriate, is considered maladaptive. This situation typically describes most perceptual speed or RT tasks. Accordingly, a large body of research investigating intraindividual variability on RT tasks has found greater variability is associated with poorer levels of performance on other cognitive tasks (e.g.,
Hultsch et al., 2002, Rabbitt et al., 2001). Overall, the type of task employed may play a role in the type of variability observed.
For the present studies, we were interested in the hypothesis that intraindividual variability reflects brain-based fluctuations, and will consequently be defining
intraindividual variability as moment-to-moment changes or fluctuations in performance across trials. We also used a univariate scope, and focused on the theory that
intraindividual variability is maladaptive and thus indicative of brain-based dysfunction. This type of variability has also been termed inconsistency (Hultsch et al., 2000) and both terms will be used interchangeably throughout this dissertation.
Reliability of Intraindividual Variability
Despite the varying definitions of intraindividual variability, and arguments about its essential role in the developmental process, what information can be gained by
studying inconsistency in performance? Fluctuations around a person’s mean performance are usually deemed to only represent error in responding because inconsistency appears to be random. But are there any logical patterns within intraindividual variability in performance?
In fact, even within a concept representing change and development, research has revealed that there is considerable stability in inconsistency. That is, the amount of fluctuation in cognitive performance has been shown to be a relatively stable characteristic of an individual. This evidence can be divided into four domains: the stability of intraindividual variability, its relationship to other stable characteristics, variability within stable characteristics, and the neurological correlates of intraindividual variability.
A number of studies have compared the amount of intraindividual variability present on different occasions of the same task, and across similar but slightly different tasks. First, if inconsistency is a stable trait of an individual, the amount of fluctuation in performance at one point in time should be positively correlated with the amount of fluctuation in performance at another point in time, and also with the amount of fluctuation seen on similar tasks. On the other hand, if inconsistency in cognitive performance is completely random, the measurements should be unrelated. In a comparison including healthy, arthritic, and demented older participants, Hultsch et al. (2000) found that individuals who were more variable across tasks on one occasion, also tended to be more variable across tasks on other occasions, and those who were more variable trial-to-trial on one RT task, were also more variable trial-to-trial on other RT tasks. Similar findings were found by Fuentes, Hunter, Strauss, and Hultsch (2001), Hultsch and colleagues (2002), Rabbitt and colleagues (2001), and Allaire and Marsiske (2005). Further, positive correlations have even been found for comparisons which computed separate reliability measures for odd and even trials (e.g., Jensen, 1992) or split the occasions in half and calculated two reliability statistics (Eizenman et al., 1997).
Next, if inconsistency in cognitive performance is characteristic to an individual, the amount of variability on one cognitive task should be positively correlated with the amount of variability on another cognitive task. The positive link would be expected if intraindividual variability were the cause of relatively stable endogenous mechanisms such as neurological dysfunction that were present during the performance of each task, rather than relatively transient exogenous influences such as fatigue or stress that may be present during one task and not the other. Research results have been consistent with this interpretation, and found individuals who showed more fluctuation in performance on one
RT task tended to show more fluctuation on other RT tasks (Hultsch et al., 2000; 2002; Fuentes et al., 2001). Relatedly, Hultsch and colleagues (2002) found a significant positive relationship between two types of intraindividual variability: those with greater dispersion or intraindividual variability in performance across tasks tended to also show greater inconsistency across trials on all tasks.
Finally, regardless of the stability of intraindividual variability in cognitive performance, is it of sufficient magnitude to warrant interest and study, particularly relative to other types of variability? Nesselroade and Salthouse (2004) argued that if intraindividual variability is small relative to interindividual differences, the phenomenon may only be of theoretical interest. On the other hand, if intraindividual variability is of substantial size, and not simply error variance, its presence demands an explanation. Nesselroade and Salthouse (2004) investigated this issue by comparing three types of variability in adults’ performance on perceptual-motor tasks: a) between-person variability; b) within-person variability across trials of the task; and c) within-person variability across three occasions of the task. They found that the size of intraindividual variability was about one half as large as that of interindividual variability, a conclusion consistent with later findings on accuracy tasks (Salthouse, Nesselroade, & Berish, 2006), and memory and sensorimotor domains (Li, Aggen, Nesselroade, & Baltes, 2001). Further, by manipulating the regression equation, Nesselroade and Salthouse (2004) inferred that the average within-person variability for that particular task was
approximately equal to the variation expected to occur across a 26-year period of normal aging. Ram, Rabbitt, Stollery, and Nesselroade (2005) also found that there were
substantial individual differences in intraindividual variability in intraindividual change across 36 occasions of a cognitive task. Individuals varied in how consistent they were in
responding at the first occasion of the task, their change in intraindividual variability across the occasions (e.g., some individuals improved more quickly than others), and the final level of inconsistency in performing the RT task, further demonstrating the
individual nature of intraindividual variability. Clearly, inconsistency is of significant magnitude and stability to warrant interest in development.
Intraindividual Variability and Various Stable Characteristics
A critical test of intraindividual variability is whether it is related to stable individual characteristics, which could only occur if there was some constancy to the pattern of intraindividual variability displayed. Further, the relationship would need to occur in a way consistent with underlying hypotheses and theory. For example, if inconsistency is believed to be an indicator of maladaptive development, the relationship between adaptive developmental outcomes such as high cognitive ability and
inconsistency should be negative (i.e., higher test score, lower inconsistency in performance). In fact, there have been numerous studies demonstrating significant relationships between intraindividual variability and an impressive range of ability and theoretical domains.
The first and most pronounced correlate of inconsistency in performance is with intelligence and cognitive ability. Early studies noted negative correlations of
intraindividual variability with general intelligence, in that low aptitude individuals were excessively variable from one RT trial to the next compared to brighter individuals (Jensen, 1982; 1992). The expected relationship with IQ was also evident for both within-session and across-session variability (Rabbitt et al., 2001). In fact, a later growth curve analysis on the same data showed that those with higher intelligence reached lower asymptotic levels of inconsistency with practice on the RT task (i.e., they reached higher
levels of proficiency as evidenced by their greater consistency in responding), and
regardless of practice effects, did not vary as much week-to-week compared to those with lower scores (Ram et al., 2005). Similar findings have been found in relation to cognitive tasks covering a range of cognitive domains. Hultsch and colleagues (2002) found more inconsistent performance on four different RT tasks was associated with lower scores on tasks assessing perceptual speed, working memory, episodic memory, and crystallized abilities. Greater inconsistency in RT performance was even predictive of poorer
everyday problem solving abilities, such as the ability to decipher nutritional information or transportation cost (Burton, Strauss, Hunter, & Hultsch, in press).
The negative relationship with inconsistency in cognitive performance has even extended beyond performance indicators such as cognitive ability to include adaptive lifestyle behaviours. Bielak, Hughes, Small, and Dixon (2007) found older adults who had a higher frequency of participating in leisure activities such as completing
crosswords, using the computer, or playing a musical instrument demonstrated less intraindividual variability on four RT tasks. Therefore, the cognitive benefits of maintaining an active lifestyle, or the “use it or lose it” hypothesis of cognitive aging were even evident in older individual’s fluctuations in their cognitive performance. Further, inconsistency at baseline testing was predictive of later attrition in a 6-year longitudinal study of aging. MacDonald, Hultsch, and Dixon (2003) found significant differences between returnees and dropouts on simple, lexical decision, and semantic decision RT tasks. Individuals who discontinued their participation at Wave 1 were more inconsistent in responding than those who returned for the next two waves of testing. MacDonald and colleagues noted that it is well-known that attrition is not random, but tends to reflect underlying influences such as disease and cognitive impairment.
Therefore, intraindividual variability may be a sensitive indicator of these impending problems. Clearly, intraindividual variability is a stable characteristic of an individual, but higher amounts of inconsistency are maladaptive, as evidenced by poorer cognition, poorer health behaviours, and increased risk of drop-out, which can be an indication of a variety of underlying health problems (e.g., deteriorating health, abnormal aging and disease, or impending death). In fact, a recent study by MacDonald, Hultsch, and Dixon (in press) verified these conclusions by finding that inconsistency significantly increased per additional year closer to death.
Intraindividual Variability in Other Domains
Research has shown that significant amounts of intraindividual variability also exist in other functioning domains presumed to stable. For example, Li and colleagues (2001) asked older adults to complete three walking tasks biweekly for 7 months (i.e., turn 360 degrees, walk 10 feet at a normal pace, and walk 10 feet at a fast pace). They found significant within-person fluctuation in older adults’ sensorimotor performance; the magnitude was approximately half the magnitude of interindividual differences.
Furthermore, sensorimotor inconsistency was negatively correlated with level of performance on the walking tasks (i.e., longer time, more steps to turn, and slower walking pace), and text and spatial memory. Sensorimotor inconsistency was of similar magnitude, and reflected the same pattern of negative relationships that have been shown with cognitive inconsistency (e.g., Nesselroade & Salthouse, 2004; Hultsch et al., 2000), indicating inconsistency may be an stable individual attribute that influences performance regardless of ability domain.
Further evidence for this conclusion has been found in the domain of self-reported beliefs. A seminal study by Eizenman et al. (1997) assessed the week-to-week
fluctuation in older adults’ perceived competence and locus of control. They found substantial variability in the weekly measurements, but the amount of variability was a stable individual difference as substantiated by high correlations among the beliefs indices when the 25 occasions were split in half. Moreover, Eizenman and colleagues found high perceived control did not predict mortality 5.5 years later, but the consistency in one’s control beliefs did! Those who were more variable on the control measures had an increased likelihood of dying 5.5 years later. Therefore, the trait of inconsistency appears to be indicative of a vital underlying endogenous mechanism. Bielak, Hultsch, Levy-Ajzenkopf et al. (2007) also found significant intraindividual variability in older adults’ reported general perceived control beliefs (i.e., locus of control) and memory-specific control beliefs (i.e., ways of improving their memory) across 10 bimonthly occasions. Despite the fact that participants completed these measures both before and after completing a battery of cognitive tasks, there was little evidence that these short-term changes were driven by changes in their actual performance.
Finally, some researchers have investigated the link between the variability found in one domain of performance (i.e., cognitive) with that found in another domain (i.e., physical, self-perceived affect/beliefs). Strauss, MacDonald, Hunter, Moll, and Hultsch (2002) hypothesized that if intraindividual variability was caused by endogenous
mechanisms, the fluctuations in performing a cognitive task would be related to fluctuations in performing other domains. Specifically, inconsistency in physical performance (e.g., blood pressure, respiratory function) would correspond to
inconsistency in cognitive performance if the same mechanism (i.e., brain-based) was the underlying cause of both expressions of variability. Therefore, just like cognitive
ability. On the other hand, the instability in an individual’s perceptions, beliefs, and affect would show a weaker relationship with the index of cognitive inconsistency, because such variability was likely influenced by exogenous mechanisms such as stress and mood. Three different groups of older adults (i.e., those with mild dementia, arthritis, and healthy controls) completed a variety of physical (e.g., balance/gait, fine motor dexterity), affective/belief-based (e.g., positive and negative affect, perceived competence and control), and RT tasks over four weekly sessions. There was evidence of the
expected cross-domain links between inconsistency in physical functioning and
inconsistency in cognitive performance for all groups on simple cognitive tasks, but only for those with dementia on more challenging tasks. Further, increased inconsistency in non-cognitive domains was generally associated with poorer cognitive function. For example, greater intraindividual variability in diastolic blood pressure and non-dominant finger tapping was associated with poorer memory performance. However, while physical inconsistency uniquely predicted level of cognitive performance, variability in affect and perceptions did not. Overall, the positive association between cognitive and physical inconsistency, and the dissociation between the physical and affect/beliefs domains, added significant support to inconsistency’s role as a stable, endogenous characteristic of an individual.
Neurological Correlates of Intraindividual Variability
Perhaps the most striking evidence that intraindividual variability represents a stable individual characteristic is its relationships with neurological conditions. Generally speaking, individuals who have a neurological condition or disease tend to show greater amounts of inconsistency in their behavioural responding compared to healthy
variability in behaviour as “one of the most striking results produced by a lesion of the cerebral cortex.” This relationship has been explored and consistently demonstrated in a variety of clinical pathologies including blunt head injuries, neurodegenerative diseases, and preclinical or impending disease.
First, individuals who have experienced traumatic brain injuries have been shown to be more inconsistent than healthy adults on cognitive tasks (Bleiberg, Garmoe,
Halpern, Reeves, & Nadler, 1997; Collins & Long, 1996), particularly for those with greater cognitive complexity (Hetherington, Stuss, & Finlayson, 1996). Moreover, Hetherington and colleagues found individuals who were at earlier points in their
recovery process (i.e., 5 years since the incident) were more inconsistent than individuals who were farther along in their recovery (i.e., 10 years). Further, the link between head injury and intraindividual variability was particularly pronounced when the frontal brain regions were damaged (Stuss, Murphy, Binns, & Alexander, 2003). Burton, Hultsch, Strauss and Hunter (2002) also found that individuals with head injury showed greater inconsistency in physical functioning across occasions than healthy adults, further supporting the hypothesis that intraindividual variability is a stable characteristic of an individual.
The neurological disturbance does not need to be due to a discrete insult however to be behaviourally associated with greater fluctuations in cognitive functioning. Rather, an abundance of evidence has shown that individuals with diseases that cause progressive neurological deterioration are more variable in their cognitive performance. A seminal study by Hultsch and colleagues (2000) compared the intraindividual variability in responding to RT tasks in three different groups of older adults: those with clinical
twice as much inconsistency as did the healthy and arthritic adults, who did not differ. Therefore, intraindividual variability appeared to be the result of endogenous neurological conditions, rather than exogenous somatic conditions like arthritis. Similar findings were reported by Strauss et al. (2002), who found individuals with dementia were more
inconsistent in their physical functioning as well. It has since been demonstrated that the amount of intraindividual variability may vary even among neurological impairments. Burton, Strauss, Hultsch, Moll and Hunter (2006) investigated whether increased inconsistency was apparent regardless of the type of neurological condition, indicating potential causation by general neurological disturbance, or was exclusively or more strongly related to certain types of neurological disturbance, signifying variability’s closer ties to a specific type of neurological insult. They compared older individuals with Parkinson’s disease, Alzheimer’s disease, and those with no neurological condition on 4 RT tasks. Consistent with previous research, both disease groups were more inconsistent than the healthy adults on all cognitive tasks, but the Alzheimer’s individuals also
displayed more variability than the Parkinson’s patients.
Even stronger distinctions among the types of neurological disturbance have been made. Murtha, Cismaru, Waechter, and Chertkow (2002) found individuals with frontal lobe dementia showed greater amounts of inconsistency than individuals with
Alzheimer’s dementia. Relatedly, Walker et al. (2000) showed that patients with Lewy body dementia exhibited greater fluctuations in their cognition and attention than those with vascular or Alzheimer’s dementia. Lewy body and frontal lobe dementia cause greater deterioration of the frontal lobes than Alzheimer’s dementia (Hultsch et al., 2008), suggesting that inconsistency is predominantly linked to insult in the frontal brain
regions. Such a hypothesis fits with suggestions from the brain injury literature (Stuss et al., 2003).
Finally, given the demonstrations of increased intraindividual variability among those with neurodegenerative diseases, it has recently been investigated whether
increased intraindividual variability may act as an early predictor of impending disease such as dementia. Mild cognitive impairment (MCI; Petersen et al., 1999) is proposed as an intermediary stage between normal functioning and dementia, where individuals perform below their age- and education-matched peers, but maintain otherwise normal functioning. This is hypothesized to represent a very early stage of impending dementia, at least in some cases (e.g., Albert, 2008). Thus far, all studies investigating this
association have found that older adults with some form of MCI were more inconsistent than healthy older adults (Christensen et al., 2005; Dixon et al., 2007; Strauss, Bielak, Bunce, Hunter, & Hultsch, 2007). Further, individuals with multiple domains of cognitive impairment showed greater fluctuations in responding to cognitively
challenging RT tasks than those with an isolated (single) area of impairment (Dixon et al., 2007; Strauss et al., 2007). Therefore, not only is increased intraindividual variability related to possible preclinical neurological disease, but it can also differentiate among subtle differences in neurological functioning.
Clearly, the neurological conditions described above are relatively stable characteristics of individuals, and their consistent positive links with inconsistency support the notion of reliability in within-person fluctuations in performance. The congruent findings between neurological conditions and increased intraindividual variability have also been key pieces of evidence regarding the hypothesis that
integrity. Specifically, increased inconsistency in performance appears to be indicative of brain disturbance or dysfunction (e.g., Hultsch & MacDonald, 2004; Li & Lindenberger, 1999). Further, the above research supports the association between intraindividual variability and the severity of neurological disturbance, where greater amounts of variability were observed with more severe brain disturbances, particularly those that affected frontal brain regions (e.g., Murtha et al., 2002; Stuss et al., 2003). Therefore, the degree to which an individual is inconsistent in their cognitive responding relative to healthy individuals may be a reliable measure of their neurological integrity.
Although the exact neurological cause of increased intraindividual variability is unknown, there are several possible determinants (MacDonald, Nyberg, & Bäckman, 2006). MacDonald and colleagues noted structural causes such as lesions to frontal gray matter, problems with neuronal connectivity because of white matter demyelization, and possible neuromodulatory changes. Li and Lindenberger (1999) showed that changing the gain parameter of catecholamines (neurotransmitters which affect the responsivity of neurons to incoming signals, effectively changing the neuronal signal-to-noise ratio) in neurocomputational simulations changed the resulting amount of inconsistency in cognitive performance. Most importantly, decreases in the gain parameter of the neurotransmitter, or decreased responsivity to neuronal signals, resulted in increased variability.
Despite the apparent viability of the hypothesis that intraindividual variability is a sensitive marker of neurological integrity, the above research is based on correlational and indirect associations with neurological disorders. Recently however, this hypothesis has shown further confirmation from brain-based research, and added to the verification of inconsistency’s individual stability.
If intraindividual variability truly is a marker of neurological integrity, research that directly assesses the structural characteristics and activation of the brain should show significant associations with behavioural inconsistency. Further, the relationship should be in the expected direction and consistent with the hypothesis that increased
intraindividual variability is maladaptive; thus, greater behavioural fluctuation in performance, more structural damage and maladaptive activation patterns. Although there are few studies that have investigated this association thus far, the evidence is positive. First, using functional magnetic resonance imaging (fMRI) on young to middle aged adults, Bellgrove, Hester, and Garavan (2004) found a significant association between intraindividual variability on a go/no-go response task and activation of the frontal cortex. As expected, increased inconsistency on the task was associated with poorer performance, which translated into poorer inhibition of the “go” response.
However, rather than decreased brain activation, poorer and more inconsistent performers elicited more frontal brain activity in order to inhibit the target behaviour in the “no-go” situation. Next, studies have found significant links with structural brain characteristics as well. Bunce and colleagues (2007) completed fMRI scans on healthy older adults aged 60-64 years. They found frontal lobe white matter hyperintensities (WMH; i.e., white matter lesions that affect the efficiency of neuronal conduction) were significantly associated with intraindividual variability on speeded cognitive tasks. More importantly, this relationship was unique to WMHs in the frontal lobes, clearly demonstrating that deterioration of neural pathways in the frontal cortex plays a key role in increased intraindividual variability. However, although the association was significant, the resulting effect size and strength of the relationship appeared to be minimal. This was possibly due to the relatively good health of the participant sample. The final study by
Anstey et al. (2007) supports that possibility. Anstey et al. found only minimal
associations between the size of the corpus callosum (CC) as detected by MRI scans and inconsistency on simple and choice reaction time in a healthy older sample, but
significantly stronger relationships for those with a mild cognitive disorder (i.e., MCI; age-associated memory impairment or cognitive decline; mild neurocognitive disorder). Despite the inherent inclination to then propose that the mild cognitive disorder sample had more problems in the brain (in this case, a smaller CC area), and thus a greater link with inconsistency, there was no difference in the average CC size between the two groups. However, this finding can still be in accordance with the idea that intraindividual variability is an indicator of neurological disturbance. Anstey and colleagues proposed that the biological limits of brain reserve capacity must have been reached for the mild cognitive disorder group, or that they were functioning at their maximum ability, resulting in their stronger brain-behaviour relationship.
Although the association between brain structures and activation with behavioural inconsistency is not perfect or particularly strong, the significant associations serve as direct evidence that inconsistency in behaviour is linked to brain-based characteristics. In fact, Bunce and colleagues (2007) noted that it is likely the combination of a number of underlying neurological factors, including declines in catecholamines and structural changes that contribute to increased inconsistency. For example, Kelly, Uddin, Biswal, Castellanos, and Milham (2008) investigated the possibility that when completing a demanding cognitive task, individuals need to suppress the “default mode” network of brain regions which show activity when the brain is at rest (e.g., sleep) in addition to activating those required to complete the task. They hypothesized that poor suppression of the default mode network resulted in poor task performance and increased variability.
Using event-related fMRI scans, they computed the strength of individuals’ negative correlation between their default mode and their task-positive (i.e., on task) networks, serving as an index of the degree of regulation of activity in those networks (i.e., a weak negative correlation indicated a greater likelihood that the two networks were
simultaneously active whereas a strong negative correlation indicated adequate
suppression of the default mode network while the task-positive network was active). Consistent with their expectations, Kelly et al. found that the strength of individuals’ negative correlations was related to individual differences in variability in performing an arrow RT task. Therefore, it may also be the case that inconsistency in cognitive
performance is related to an individual’s ability to regulate competing neural processes, in addition to their structural characteristics and neuromodulatory mechanisms.
Overall, a substantial amount of research demonstrates that intraindividual variability in cognitive performance is a stable, endogenous characteristic of an
individual. Inconsistency is of significant magnitude and stability relative to other types of variability (e.g., Hultsch et al., 2000; Nesselroade & Salthouse, 2004; Rabbitt et al., 2001), significantly related to level of ability in a variety of cognitive domains (e.g., Burton et al., in press; Hultsch et al., 2002; Jensen, 1992), and correlated with individual characteristics such as activity participation (Bielak, Hughes et al., 2007), risk of attrition (MacDonald et al., 2003), and even impending death (MacDonald et al., in press). An individual’s inconsistency in cognitive functioning is also significantly related to their amount of inconsistency in other domains of functioning, such as sensorimotor
performance (Li et al., 2001), physical ability (Strauss et al., 2002) and control beliefs (Bielak, Hultsch et al., 2007; Eizenman et al., 1997). Increased inconsistency is associated with a variety of neurological conditions, including traumatic brain injury
(e.g., Bleiberg et al., 1997; Stuss et al., 2003), dementia (e.g., Hultsch et al., 2000), Parkinson’s disease (Burton et al., 2006) and symptoms consistent with mild cognitive impairment (e.g., Christensen et al., 2005). Behavioural inconsistency is also
significantly related to brain activation (Bellgrove et al., 2004), structural brain characteristics (Bunce et al., 2007; Anstey et al., 2007), the availability of
neurotransmitters (Li & Lindenberger, 1999), and the ability to regulate competing neural processes (Kelly et al., 2008). Further, the above findings have supported the proposal that inconsistency is a sensitive indicator of central nervous system disturbance (Li & Lindenberger, 1999). Clearly, inconsistency in cognitive performance is a worthwhile area of interest in developmental research. Interestingly, inconsistency is related to the normative aging process as well.
Age Differences and Changes in Intraindividual Variability Cross-sectional Differences Across the Lifespan
A number of studies have found that intraindividual variability is notably higher at the two ends of the lifespan: childhood and older adulthood. Using cross-sectional data, Williams, Hultsch, Strauss, Hunter, and Tannock (2005) found cognitive inconsistency between the ages of 6 and 81 followed a U-shaped curve. Inconsistency was highest for the youngest age group, sharply decreased in adolescence and young adulthood, and slowly increased again through middle and older adulthood. Similar findings have been shown across the age ranges of 5-76 years (Williams, Strauss, Hultsch, & Hunter, 2007) and 6-89 years (Li, Lindenberger et al., 2004). However, the underlying mechanism of increased variability appears to be different at the two points in the lifespan: For the younger half of the sample, age accounted for a significant proportion of the variance in inconsistency in both the fast and slow ends of the RT distribution, and partialing the
effects of inconsistency in the faster RT trials partially reduced the age-related effects of inconsistency in the slow RT trials. The reverse was also true. Williams and colleagues (2005) argued that the mutual attenuation of the age-related effects by the slow and fast ends of the distribution indicated a general underlying process for inconsistency in childhood, in addition to specific influences unique to the faster and slower ends of the distribution. On the other hand, age differences for older adults were minimal in the fast end of the RT distribution relative to those in the slow end, and partialing inconsistency in the fast end of the distribution did not attenuate the age effect in the slow end. These results were consistent with proposals that inconsistency in older adulthood might be due to specific variability-producing processes such as attentional lapses (as evidenced by variability among the slower RT trials), rather than a general process such as slowing (e.g., Hultsch et al., 2002).
Reconciling these findings with the perspective that inconsistency is a marker of neurological integrity (e.g., Li & Lindenberger, 1999), it appears that as the brain develops in childhood, or is becoming differentiated and specialized, it is relatively unstable, and the greater fluctuations in children’s cognitive performance reflect that. During older adulthood however, the brain appears to develop the opposite way, and undergoes dedifferentiation, or increasing generalization of the brain structures that control specific cognitive processes. The dedifferentiation hypothesis coincides with poorer cognitive performance with increasing older age, and more compressed functional organization of intellectual abilities (Li, Lindenberger et al., 2004). However, Li,
Lindenberger and colleagues (2004) found that intraindividual variability is predictive of fluid intelligence and chronological age only in late adulthood and older age, and not in childhood. Therefore, intraindividual variability in cognitive performance may be
particularly informative of the integrity of the aging brain and impending aging-related disease. Further, if increased inconsistency suggests increased neurological disturbance, but also increases with older age, this measure may be particularly insightful in
explaining normative cognitive aging.
In fact, greater variability in cognitive responding among healthy older adults compared to younger adults is well substantiated (e.g., Anstey, Dear, Christensen, & Jorm, 2005; Bunce, MacDonald, & Hultsch, 2004; MacDonald, Hultsch, & Bunce, 2006; Nesselroade & Salthouse, 2004). Hultsch and colleagues (2002) compared younger adults aged 17 to 36 years with three groups of older adults: young-old (54-64 years), mid-old (65-74 years), and old-old (75-94 years). The participants completed 4 RT tasks of varying cognitive difficulty, ranging from simple RT and choice RT to lexical and semantic decision. On all 4 tasks, older adults showed more trial-to-trial inconsistency in performance than the younger adults, particularly compared to those in the oldest age group (i.e., 75 years and up). Further, differences were apparent even within the older adult age range: old-old adults demonstrated significantly more intraindividual variability in cognitive performance than young-old (i.e., 55-64 years), and mid-old (i.e., 65-74 years) adults.
Interestingly, age differences in intraindividual variability are most apparent on speeded tasks that challenge cognitive functioning, such as those that place large demands on executive functioning or working memory. For example, Dixon and colleagues (2007) found significant age differences on all 3 RT tasks (simple RT, choice RT, and 1-back choice RT), but the greatest age effects occurred on the 1-back choice RT which required participants to ignore the present stimulus and instead respond to the target stimulus presented on the previous trial. Similar findings have been found for neurological
conditions, where the greatest distinctions among groups were on cognitively complex tasks (e.g., dementia versus healthy, Hultsch et al., 2000; dementia subtypes, Murtha et al., 2002; CIND status, Strauss et al., 2007). However, it does not appear to be the case that the complex cognitive tasks simply exacerbate the effects of the poorer motor and perceptual functioning which accompany normal older age. Bunce et al. (2004) demonstrated that greater variability in older adulthood was not due to slower motor processing, but rather caused by increased attentional and cognitive demands. Further, MacDonald, Hultsch et al. (2006) investigated whether decreased perceptual functioning in older age was the cause of greater inconsistency on complex RT tasks by asking if younger adults could be experimentally manipulated to show inconsistency levels on par with adults who were decades older. Visual degradation of the stimuli however, did not successfully age younger adults’ performance to mimic that of older adults’, supporting the proposal that greater inconsistency with age is due to a more general endogenous process.
Longitudinal Changes with Age
However, any true test of a developmental phenomenon requires longitudinal evidence. Because the above studies were cross-sectional, they only demonstrate that there are age differences in intraindividual variability in older adulthood, and not that intraindividual variability increases with older age, as such change statements require longitudinal data. To date, relatively few studies focusing on inconsistency have
employed longitudinal techniques, and only two have investigated its relationship to other meaningful outcomes. However, the results are promising.
First, several studies have demonstrated that intraindividual variability does in fact increase with older age. An 8-year study by Fozard, Vercruyssen, Reynolds, Hancock,
and Quilter (1994) found within-person variability in responding to audio RT tasks increased over time, particularly for the more challenging task. A more thorough population-based study by Deary and Der (2005) compared the initial variability in RT performance of younger to middle-aged adults (16-63 years) to their performance 8 years later. Interestingly, they found variability in performance on simple RT remained stable up until around 50 years. In contrast, intraindividual variability on the choice RT task increased steadily from the mid-30s to the mid-60s. Therefore, in accordance with cross-sectional evidence (e.g., Strauss et al., 2007), complex RT tasks appear to be most attuned to subtle neurological changes that occur with age. Der and Deary (2006) replicated these findings with a different population-based study that spanned a much larger age range (18-94 years). Again, variability in performing the simple RT task did not increase until age 50, whereas fluctuations in performance on the complex RT tasks increased
throughout the adult age range. Both studies also found women were more inconsistent than men on the choice RT task in mid-adulthood (i.e., 36-63; Deary & Der, 2005), and across the lifespan (Der & Deary, 2006), but this finding has been questioned. Reimers and Maylor (2006) examined RTs on a trial-to-trial basis (compared to the overall SD used by Der and Deary), and found the gender effect disappeared when the first two trials of the task were removed. That is, women were initially slower than men, but became faster than men across the remaining trials.
Next, two recent studies have not only corroborated that individuals become more inconsistent as they age, but also demonstrated the utility of intraindividual variability in predicting cognitive change. MacDonald and colleagues (2003) examined 6-year
longitudinal change in inconsistency, and expanded their investigation to include whether intraindividual variability predicted corresponding change in level of cognitive
performance. Participants from the Victoria Longitudinal Study aged from 55 to 89 years (N=446) completed 4 RT tasks in addition to cognitive tests targeting processing speed, working memory, fluid reasoning, episodic memory, and crystallized verbal ability. Each test was completed at each of three waves over a 6-year period. First, there were marked increases in intraindividual variability, but only for the old-old participants (i.e., 75-94 years). The young-old (55-64 years) and mid-old (65-74 years) adults remained constant in their within-person variability or declined slightly over the 6 years. Therefore, in contrast to earlier findings (e.g., Der & Deary, 2006), inconsistency did not increase uniformly across the older age range. However, differences within the older adult range are consistent with cross-sectional findings that inconsistency is greater in the latter half of older adulthood (e.g., Hultsch et al., 2002). Further, 6-year change in inconsistency was greater on the verbal (i.e., lexical and semantic decision) compared to non-verbal RT tasks (i.e., simple and choice RT), which were arguably more cognitively complex and challenging. Next, MacDonald and colleagues investigated the longitudinal links
between inconsistency and cognitive change. Impressively, the amount of inconsistency at the initial test occasion significantly attenuated 6-year decline on the cognitive tests by an average of 93%. In other words, cognitive inconsistency at time 1 significantly predicted cognitive change over the 6 years. However, do the changes in inconsistency also covary with the changes in cognition? Using hierarchical linear modeling,
MacDonald and colleagues found significant associations between 6-year change in intraindividual variability and 6-year change in cognitive test score. On occasions where individuals were more inconsistent, they also tended to score lower on cognitive tasks completed at that wave of testing, independent of the average linear trend across time. The relationship between increasing intraindividual variability over time and declining
cognitive performance was significant for five out of the six cognitive measures (all except crystallized verbal ability). Interestingly, the significant covariation relationship was invariant across age groups, indicating the amount of cognitive decline per unit increase in inconsistency did not vary as a function of age. Therefore, although increases in inconsistency tended to be larger in old-old adulthood, the relationship between
inconsistency and cognition remained stable across the older age range. These findings fit corresponding larger declines in cognitive ability past 75 years of age.
Lövdén, Li, Shing, and Lindenberger (2007) conducted a similar investigation with the old and very old (70-102 years) from the population-based Berlin Aging Study. They wanted to replicate the findings of MacDonald et al. (2003) with a much older sample and with five occasions of testing across a longer time frame (i.e., 13 years). They also wanted to extend these results by investigating the lead-lag relationship between inconsistency and cognition. Participants’ level of cognitive ability was based on performance in two different cognitive domains: perceptual speed (assessed by digit letter) and ideational fluency (assessed by categories). Intraindividual variability was derived from another perceptual speed task (identical pictures) that recorded RT for each trial. Using latent growth curve modeling, Lövdén et al. found results consistent with those reported by MacDonald et al. (2003): individuals across the age range (70-102 years) became more inconsistent over time, with an average 1.2 unit increase every 2 years. The uniform increase in variability across the sample is consistent with the findings for the old-old group (75-94 years) in MacDonald et al. Lövdén and colleagues also found significant associations between longitudinal change in inconsistency and longitudinal changes in ideational fluency and perceptual speed, such that individuals who were more variable at one testing occasion tended to also score lower on the
cognitive tasks at that occasion. Further, the covariation relationship was significant across the entire range from 70-102 years, corroborating that the link between
inconsistency and cognition remained stable in older adulthood (i.e., MacDonald et al., 2003). Overall, Lövdén and colleagues confirmed that significant covariation between cognition and inconsistency was apparent even among the old and very old in a
population-based study, and more impressively, these results were not confounded by chronological age, suspected dementia, or time-to-death.
However, because random effects models are limited in their ability to address lead-lag relations as change and level are defined over the same period, and thus level does not precede change, it was unknown whether one variable caused change in the other. Did inconsistency signal subsequent negative longitudinal changes in cognitive performance, or did the relationship operate in the opposite direction, where cognitive decline occurred before participants showed more behavioural variability? To examine this further, Lövdén and colleagues employed bivariate dual change score models, which allowed for corresponding time-lagged associations between the variables and direct and simultaneous modeling of the variables’ changes. The results showed greater
inconsistency reliably preceded and predicted greater 2-year negative decline in ideational fluency (categories) and perceptual speed (digit letter), indicating that intraindividual variability could predict looming changes in level of cognitive performance. Further, the opposite model of level of ideational fluency did not influence later change in
intraindividual variability. However, this same dissociation did not hold across cognitive domains; greater perceptual speed also predicted greater 2-year increased change in inconsistency. Regardless, in the first study to examine the lead-lag relationship between inconsistency in cognitive performance and cognitive level, there was evidence that
inconsistency may be “a developmentally early flag for impending old-age changes in mean levels of cognitive performance” (p. 2834), even if the strength of this predictive relationship might vary across cognitive domains. Further, this predictive possibility is particularly promising because the time-lagged relationship was apparent regardless of chronological age, suspected dementia, and time-to-death.
Overall, it is well documented that intraindividual variability is significantly associated with the aging process. Inconsistency in cognitive performance is positively associated with age, with older adults being more inconsistent than younger adults
(Hultsch et al., 2008), and also significantly increasing with age, beginning around age 70 (e.g., Lövdén et al., 2007; MacDonald et al., 2003). More importantly, longitudinal studies of aging have verified the link between greater inconsistency and poorer cognitive ability over time by finding significant coupling relationships between change in
intraindividual variability and change in cognition across 6 (MacDonald et al., 2003) and 13 years (Lövdén et al., 2007). Finally, Lövdén and colleagues (2007) identified the temporal order of this relationship among the old and very old adults as change in
inconsistency preceding change in cognition, indicating that intraindividual variability in older adulthood may be a sensitive marker of impending cognitive decline or disease. This dissertation built on these findings by delving deeper into age-related change in intraindividual variability, its links within cognitive change, and whether intraindividual variability truly is indicative of later developmental outcomes.
Overview of Studies
A main goal of developmental research is to evaluate whether current outcomes can be predicted by certain preceding factors or variables. These types of studies are particularly valuable and informative of the developmental process, but few such studies
exist in the intraindividual variability literature. Longitudinal data is essential for this research area because intraindividual variability is believed to be an early behavioural marker of neurological disturbance and impending cognitive decline. Therefore, the only way to validate this hypothesis is through prospective longitudinal investigations. This dissertation presents two studies of intraindividual variability in a longitudinal context to further explore the relationship between short-term intraindividual variability and longer-term cognitive change in older adults. The first study addressed change in inconsistency over time, and the covariation of change in inconsistency with change in cognition. The second study addressed the capacity of intraindividual variability to predict later cognitive ability and other meaningful change outcomes such as cognitive status and attrition.
Study 1
There is substantial evidence that trial-to-trial fluctuations in cognitive performance increase as a result of the aging process. Specifically, intraindividual
variability appears to linearly increase from early adulthood well into late older adulthood (e.g., mid-30s to mid-60s, Deary & Der, 2005; 18-94 years, Der & Deary, 2006), with an increased acceleration in old-old adulthood (e.g., 70-102 years, Lövdén et al., 2007; 75-89 years, MacDonald et al., 2003). Interestingly however, greater increases in
intraindividual variability in old-old adulthood do not appear to change the underlying coupling relationship with cognitive performance. Rather, the relationship between cognition and variability appears to remain stable across old adulthood (i.e., 55-89 years, MacDonald et al., 2003), where the amount of cognitive decline per unit increase in inconsistency does not vary by age. Further, Lövdén et al. (2007) presented the first temporal evidence that inconsistency in cognitive performance significantly precedes cognitive decline, and may thus be able to predict impending cognitive changes in older
adulthood. However, because only two studies thus far have investigated the covariation relationship between inconsistency and cognition in older adulthood (i.e., Lövdén et al., 2007; MacDonald et al., 2003), it is unknown if the relationship will generalize to other time scales or age groups. Therefore, the longitudinal link between inconsistency and cognitive performance demands further replication and extension (Lövdén et al., 2007; Hultsch & MacDonald, 2004).
The age-related increase in inconsistency also appears to vary according to the cognitive difficulty of the task. For example, Deary and Der (2005) and Der and Deary (2006) found variability in responding to a simple RT task remained stable until age 50, whereas inconsistency in completing a choice RT task increased throughout the adult age range. Further, MacDonald et al. (2003) found 6-year change in inconsistency was greater on verbal (i.e., lexical and semantic decision) compared to non-verbal RT tasks (i.e., simple and choice RT). Similar findings have been found throughout the
inconsistency literature, including inconsistency based on complex tasks providing greater sensitivity to cross-sectional age differences (e.g., Bunce et al., 2004; West, Murphy, Armilio, Craik, & Stuss, 2002), and other stable characteristics such as level and change of activity participation (Bielak, Hughes et al., 2007). Further, cognitively
demanding tasks provided the best discrimination among varying degrees of mild cognitive impairment (Dixon et al., 2007; Strauss et al., 2007; but see Christensen et al., 2005). Therefore, it appears that the variability in responding while under high cognitive demand, compared to that observed while performing simple cognitive tasks, may be most attuned to the integrity of the neurological system, and thus subtle changes that occur with age. It remains to be seen however whether the difficulty of the inconsistency-based task affects the coupling relationship with cognitive performance as well.
The present study enhanced the inconsistency literature by addressing these issues. First, in terms of the generalizabilty of the longitudinal relationships, the sample was recruited on the basis of having concerns about their memory (rather than a
normative aging study as used by MacDonald et al.), and covered the mid-old to old-old age range (i.e., 64-92 years). A shorter testing interval (i.e., yearly measurements) over 3 years was also used. Although MacDonald and colleagues used 6-year longitudinal data, because their participants were tested every three years, their covariation results were based on change relationships that occurred at 3-year intervals. Similarly, Lövdén et al. assessed the change relationship over 13 years, but because the participants were retested approximately every 2 years, the coupling of cognition and inconsistency was
investigated at 2-year intervals. The present study was conducted over 3 years, and included 4 annual testing occasions. This shorter longitudinal interval will further explore the sensitivity of intraindividual variability to short-term cognitive changes, as it may be the case that the link between cognition and variability is stable only across time periods of certain length. Can inconsistency identify even annual developmental change? Finally, the present study used 10 RT tasks of varying complexity to assess whether the longitudinal relationship between intraindividual variability and cognition is stronger for tasks requiring greater cognitive effort. Lövdén et al. used only one type of trial-to-trial variability based on completing an identical pictures task (which is a form of choice RT) and there were significant associations with both assessed cognitive domains (perceptual speed and ideational fluency). On the other hand, MacDonald et al. analyzed 4 different measures of inconsistency (i.e., 2 non-verbal and 2 verbal RT tasks) with 6 different cognitive domains (e.g., working memory, fluid reasoning, episodic memory). The covariation relationship between inconsistency and cognition was significant in each
